Knowledge externalities and demand pull:
The European evidence
Antonelli Cristiano, Università di Torino, Departimento di Economia e Statistica “Cognetti de Martiis”,
Lungo Dora Siena 100A , 10154 Turin and Collegio Carlo Alberto, BRICK, via Real Collegio 30, 10024
Moncalieri (Turin), Italy
and
Gehringer Agnieszka, Georg-August Universität Göttingen, Lehrstuhl für Wirtschaftspolitik, Platz der
Göttinger Sieben 3, 37073 Göttingen, Germany
Abstract
This paper elaborates the microeconomic foundations of the demand pull hypothesis stressing the role of vertical
knowledge externalities stemming from user-producer knowledge interactions that parallel vertical transactions. The
paper articulates and tests the hypothesis that such competent demand is actually able to pull technological change only
when it is expressed by advanced users, able to provide relevant knowledge externalities to their customers. Using input
output tables we test empirically this hypothesis for 15 European countries in the years 1995-2007. The evidence
confirms that the increase in productivity of the upstream sectors is positively influenced by the sector-level derived
demand, according to the upstream rates of introduction of innovations and to the intensity of the user-producer
interactions. The policy implications of the analysis enable to elaborate and implement the notion of a ‘competent’
public demand.
JEL classifications: L16; O3; O52
Keywords: Derived demand; Knowledge externalities; User-producer sectoral interactions; Input output tables
1
1. Introduction
The positive effects of demand on innovation-led economic growth have been advocated for decades with little
empirical support and increasing theoretical skepticism. The advances of the new economics of knowledge and the new
understanding of technological change as an emergent system property enable to articulate the foundations of the
demand pull hypothesis in a novel interpretative context that stresses the central role of external knowledge spilling
from innovative customers in supporting the arrival of innovations generated by the upstream producers. Thus, in this
novel view, demand is not generic but competent.
The original Kaldorian demand pull hypothesis was elaborated in macroeconomic analytical framework,
strongly focusing on increasing returns from public intervention to support aggregate demand. Complementarily to the
analysis of Kaldor, the later work by Schmookler provided additional clearness in explaining the causal chain of effects
linking the increase in aggregate demand to the positive impact on investment and, subsequently, on technological
advance. In the analysis of Schmookler, not a generic demand pull, but rather the specific effects of the derived
demand, originating from private investment in particular sectors, play a substantial role in shaping the future
technological development, crucially based on effective and continuous transactions between upstream producers and
downstream users.
Nevertheless, none of these contributions has yet taken advantage of the recent advances of the economics of
knowledge and specifically of the crucial role of knowledge externalities in the generation of new knowledge to
elaborate a rigorous microeconomic analysis of the mechanisms that relate the increase of demand at the aggregate level
with the increase in the rate of introduction of innovations. The present contribution aims to fill this gap.
Past empirical investigations around the standard demand pull hypothesis delivered the evidence of a positive
influence of investment-sustained derived demand on upstream innovativeness, stressing the role of upstream
profitability in providing the necessary incentive to innovate. Surprisingly, little empirical investigation implemented
input output analysis, which, instead, offers detailed information about the relations among sectors along the vertical
chain of production. In our empirical investigation, in which we test the role of the competent demand in spurring
upstream technological advances, we implement, thus, input output tables to measure the intensity of intersectoral
transactions based on knowledge interactions.
On the conceptual side, our contribution aims at revisiting the original demand pull hypothesis by integrating the
macroeconomic approach with a microeconomic analysis, which elaborates upon the central role of user-producertransactions-cum-interactions as carriers of knowledge externalities. In this spirit, the derived demand of downstream
users exerts a double effect on the innovativeness of upstream producers: i) it increases the demand for the upstream
2
products with positive effects on profitability and investments and ii) it provides influential interactions parallel to
vertical transaction flows, in which innovative downstream users support upstream innovative activity with the
provision of external knowledge that is stronger the stronger their technological advance. This is to say that upstream
innovations benefit not only from positive profit incentive, but also and crucially from knowledge interactions with the
users of their products. In such a system of vertical transactions-cum-interactions, the stronger are the knowledge
externalities, the more efficient is the expected outcome both for the system as a whole and for each single node of the
vertical chain. Upstream producers can benefit from the technological advance of their downstream customers
according to the actual amount of transactions: technological externalities do not flow in the air but require systematic
interactions. Hence, the actual access to knowledge externalities is strongly conditioned on market transactions-cuminteractions, in which both knowledge producers and knowledge users actively participate.
Our theoretical arguments are tested empirically on a panel of 15 EU countries in the period 1995-2007. We
assess which role inter-sectoral influences based on productivity growth played in the innovation-driven growth
process. The results support the hypothesis that productivity-based user-producer transactions-cum-interactions were
crucial in determining the innovative dynamics of the industrial systems in Europe. In particular, upstream innovative
outcomes were not driven by generic demand-side influence, but by transactions-cum-interactions with the downstream
innovative customers and users who directed their derived and competent demand for intermediate inputs towards
upstream sectors. The stronger is technological advance and derived demand of downstream sectors and the larger the
chances that upstream industries can rely on the access to external knowledge to introduce new technologies in turn.
The rest of the paper is structured as follows. Section 2 recalls the foundation of the competent demand pull
hypothesis. Section 3 presents the main econometric analysis. The conclusions in the last section summarize the results,
emphasizing their European dimension and providing crucial though preliminary implications for the European
industrial development.
2. Theoretical framework
2.1. Demand pull hypothesis in the standard view
The demand pull approach, according to which innovation is driven by the increase of demand, has received very little
attention in the recent literature. This is consistent with the mismatch between the macroeconomic foundations of the
demand pull hypothesis and the strong microeconomic flavor of the literature on the economics of innovation.
The first specification of the demand pull hypothesis was articulated by Nicholas Kaldor who made explicit the
macroeconomic and post-Keynesian framework of analysis (Kaldor, 1966 and 1972). Kaldor suggested that all
increases of public expenditures able to support the expansion of aggregate demand would have –also- positive effects
3
on output via the interaction between the multiplier and the accelerator. Eventually, increases in investments could be
also achieved. Additional flows of investments were expected to fasten the diffusion of technological innovations
embodied in new vintages of capital goods. Hence, the eventual increase in efficiency of the system would follow,
leading ultimately to the increase of output and, consequently, of the flows of fiscal receipts to repay the original
spending. Actually, the Kaldorian approach to demand pull did focus more on the positive effects of fixed investments
on the diffusion of innovations that were necessarily embodied in capital goods, rather than on the specific stimulation
of aggregate demand leading to the actual introduction of technological innovations. The main aim of Kaldor’s analysis
was to substantiate the case for increasing returns at the system level and to show the need for the continual expansion
of public expenditures. The increase of efficiency engendered by the increase of investments stemming from additional
public expenditures and the accelerated diffusion of new vintages of capital goods would in fact generate additional
fiscal receipts neutralizing the effects of transient deficits on the stock of public debt.
The sophisticated analysis of Jacob Schmookler's Invention and Economic Growth (1966) complements the
strong post-Keynesian flavor of the Kaldorian analysis. But additionally, it makes an important step forward in
articulating the causal chain of arguments that relate the increase of the aggregate demand to investment and, finally, to
the actual generation of new technological knowledge and the introduction – as distinct from absorption - of
technological innovations.
Jacob Schmookler focuses attention on the role of a highly specific derived demand, originating from private
investments in well defined sectors of the economic system, rather than on the generic increases of aggregate demand
engendered by public expenditures. His historic analysis shows how, in the US experience, technological innovation
were pulled by specific waves of investments in a sequence of activities that were strongly related, such as canals,
railroads, electric power (…). Schmookler shows how ever since the first great undertaking, with the creation of a
national system of canals, dedicated investments were able to increase the demand for upstream industries producing
intermediary inputs and capital goods and, consequently, their profitability could also stir the inventive activity of firms,
favoring their knowledge interactions with scientists and universities at large. Each of these waves of investments had
the twin effect of stirring the increase of the innovative activities in upstream sectors and of creating the conditions for
further investment. The successful wave of investments that led to the creation of a new web of canals connecting the
Great Lakes with the ports of New York City and New Orleans, made agriculture viable in the Far West and paved the
way to its inclusion in the American economy. They also favored the demand for transportation infrastructure and the
ensuing creation of a new railroad system. In turn, investments in railroad had the twin effect, on the one side, to
stimulate the introduction of technological innovations in upstream industries providing capital and intermediary inputs
to the newly establishing industry and, on the other side, to increase the derived demand for steel.
4
The demand pull hypothesis articulated by Schmookler retains the basic macroeconomic framework as it stresses
the vertical links between specific flows of investment, additional derived demand to upstream sectors and increased
levels of inventive activity. Schmookler, however, is aware of the need of microeconomic foundations and argues that
the additional derived demand stirs the profitability of firms and this in turn would explain their rates of introduction of
innovations. In so doing, Schmookler provides a crucial, though still elementary, microeconomic framework of analysis
that highlights the role of profitability.
After the founding contributions, the demand pull hypotheses received some attention by subsequent empirical
studies that confirmed the positive effects of derived demand and investments on the intensity of innovative activity. To
measure innovativeness, different indicators have been implemented: dedicated patents (Scherer, 1982), R&D
expenditures (Jaffe, 1988; Kleinknecht and Verspagen, 1990), total factor productivity (Jaffe, 1988) and labor
productivity (Crespi and Pianta, 2008). Also different aggregation levels have been applied to verify the demand pulling
impact: sector-level, firm-level (Brouwer and Kleinknecht, 1999; Piva and Vivarelli, 2007; Fontana and Guerzoni,
2008) and the case study (Guerzoni, 2010; Nemet, 2009; Walsh, 1984). The empirical evidence confirms the positive
role of demand on the intensity of innovative activity and stresses the central role of profitability as the key factor in
exploiting the crucial demand side impact. New investments feed the derived demand directed towards upstream
sectors, hence, increase their profitability and provide at the same time incentives and opportunities for the more
systematic introduction of innovations (Andersen, 2007).
Yet the demand pull hypothesis failed to gain a broader consensus. Mowery and Rosenberg (1979) raise a
critical methodological issue, arguing that the evidence provided by the case study as well as by the early econometric
investigations could not disentangle the effects of technology pushes from demand pulls. As Mowery and Rosenberg
noted, the increase in the demand is not a determinant but often and quite obviously a consequence of the introduction
of an innovation and of the corresponding reduction of –hedonic- prices. As Dosi (1982) notes, the demand pull
approach failed “to produce sufficient evidence that ‘needs expressed through market signaling’ are the prime movers
of innovative activity”. The demand pull hypothesis suffered from the missing attention and analytical grasping of the
relation between the changes in the demand levels and the working of the mechanisms of the generation of
technological knowledge and the introduction of technological innovations (Di Stefano et al., 2012).
The lack of an appropriate microeconomic basis led to persistent ambiguities and skepticisms about the role of
demand in pulling innovation activity. As Geroski and Walter note “Although the assertion that demand matters is
controversial to some, what makes understanding the relationship between demand and economic activity really
difficult is the diversity of views which exist about whether demand makes innovative activity pro or anti-cyclical, and
5
about whether changes in demand are exogenous to the process of producing and using innovations” (Geroski and
Walter, 1995:918).
2.2. The economics of knowledge and the new microeconomic foundation of the demand pull hypothesis
The new advances of the economics of knowledge make it possible to reconsider the demand pull hypothesis in a
context that stresses the relevance of user-producer transactions and interactions as carriers of knowledge externalities
and of learning capacities of the entire industrial systems. According to the conceptual developments of the economics
of knowledge, knowledge is both an input and an output. Existing knowledge is the output of a generation process and
yet is an essential, indispensable, input into the recombinant generation of further technological knowledge (Weitzman,
1996 and 1998; Arthur, 2007 and 2009). Because no agent can possess all existing knowledge, external knowledge in
turn, is strictly necessary to each agent, to generate new technological knowledge (Antonelli and Link, 2015). In this
context, knowledge interactions are necessary because of the strong tacit content of existing knowledge. Knowledge
interactions can rarely be separated from market transactions: knowledge interactions parallel and complement market
transactions.
The access to external knowledge requires systematic and repeated interactions between knowledge holders and
its users (Antonelli, 2008). Nevertheless, the strong tacit component of knowledge makes it sticky to the routines and
the organizations where it has been conceived. Its use as an intermediary input into the recombinant generation of new
knowledge requires necessarily transactions cum dedicated interactions. Such interactions are not free, yet they require
dedicated resources for their implementation, in addition to the intentional and active participation of both parties.
Pecuniary knowledge externalities are, thus, conceptually distinct from pure technological externalities (Gehringer,
2011a). The latter postulate the absence of costs connected with the knowledge-based relations (Griliches, 1979). In this
sense and in contrast with pecuniary knowledge externalities, pure technological externalities would imply interactionssine-transactions. Vertical knowledge external effects take place within the chain of relations that link users to
producers located either in the same or different industries and are typically pecuniary as they take place in a context
where knowledge interactions are strictly associated with market transactions.
This approach stresses the role of vertical knowledge externalities stemming from the user-producer interactions
that take place along with user-producer transactions (Von Hippel, 1976 and 1988). Vertical knowledge externalities
can be identified as an important input of the knowledge generation process that differ both from the intra-industrial
Marshall externalities (often referred to as MAR externalities, after the original contribution of Marshall and of
subsequent works by Kenneth Arrow and Paul Romer) and inter-industrial, horizontal, Jacob’s (1969) externalities that
typically take place in urban environments. Vertical knowledge externalities add to intra-industrial knowledge
6
externalities that take place within the same industry and to Jacob’s horizontal knowledge externalities that take place
between co-localized, but unrelated industries.
The generation of new knowledge can take place effectively only when, where and if external knowledge can be
accessed at low costs via knowledge interactions. The access to external knowledge at costs that are below equilibrium
levels, in turn, makes it possible to increase total factor productivity (Antonelli, 2013).
Within this context, it is possible to grasp the central role of the derived and competent demand expressed by
innovative customers. The positive effects of demand pull more specifically refer to the relationship between derived
demand and increased innovative intensity that takes place when innovative customers in downstream industries are
able to stir and support the innovative activity of upstream industries. Such a positive productivity dynamics is enabled
not only because of increased profitability, but also and crucially as a consequence of intensive user-producer relations
and closer participation of each industrial node to the collective innovative efforts generated along the vertical filiére of
productive activity. The user-producer relations are crucially based on transactions among sectors that provide and
receive sequential inputs to the production of final goods. Among the inputs, externally generated technological
knowledge is exchanged on the market and triggers pecuniary knowledge externalities that catalyze innovativeness of
the economic system at large (Antonelli, 2008; Gehringer, 2011a). To give an illustrative example of the way in which
pecuniary knowledge externalities enhance the efficiency-increasing chain of reactions of the upstream producers as a
consequence of a vertical interaction with their downstream users, let us consider the development of a new machine
necessary in the production of a specific kind of apparel. The textile industry, thanks to its creative capacities (which on
their own could been acquired through the successful interactions with the apparel users), arrives at an innovative
blueprint of cold-weather performance apparel. The textile industry is a customary client of the machinery industry and
engages in interactions with the upstream suppliers. The latter make it possible to specify the knowledge-intensive
needs downstream that are crucial in the subsequent development of this specific type of apparel producing machinery.
The producers of machinery clearly benefit – although at some cost of interactions, including, for instance, the
employment of qualified R&D specialists – from the creative insights delivered from their downstream users. The more
advanced are the downstream customers and users and the more sophisticated their needs, the stronger the flows of
transactions between the two sectors and the stronger the amount of knowledge externalities that benefit upstream
producers making possible the generation of additional technological knowledge and the introduction of further
technological innovations. Taking advantage from the positive pecuniary knowledge externality, the upstream
producers finally arrive at more efficient, innovative, prototype of the apparel producing machine.
7
2.3. The competent demand pull hypothesis
The integration of the advances of the new economics of technological knowledge with the tradition of the
demand pull approach enables to elaborate the competent demand pull hypothesis in the context of innovation-based
economic development.
Innovation is an emergent property of the economic system into which firms are embedded. The characteristics
of the system in terms of flows of knowledge externalities play a crucial role in assessing the actual outcome of the
reaction of myopic firms caught in out-of-equilibrium conditions by unexpected changes in both product and factor
markets. At each point in time, firms make decisions leading to the commitment of irreversible resources in the attempt
to identify the equilibrium conditions. Yet, they cannot foresee all the possible alterations that hit the market. When the
actual conditions of both product and factor markets differ from the expected ones, as they usually do, firms try to react
(Antonelli, 2008 and 2011). Following Schumpeter (1947), their reaction can be either adaptive or creative. In the
former case, firms’ reaction is locked in the existing technology, as they only move on the existing map of isoquants,
adjusting quantities to prices and changing their production techniques, but not their technologies. Instead, the reaction
of firms is creative, when they contribute to the increase of dynamic efficiency by introducing themselves new
technologies. This corresponds to the shift downwards of the existing map of isoquants and the achievement of a more
efficient production result. The reaction of firms can be actually creative when they have access to external knowledge
and, consequently, to pecuniary knowledge externalities. The latter are the key to activate the recombinant generation of
new technological knowledge and the consequent introduction of technological innovations that makes possible the
increase of total factor productivity.
This is the context into which the competent demand hypothesis applies. When the demand for the products of
the firms exceeds the production capacity, constrained by irreversible commitments concerning both capital and labor,
they may try and improve their efficiency so as to be able to produce a larger output while retaining the current levels of
inputs. This necessarily requires a reaction that can be creative when and if they can rely upon knowledge externalities.
The chance that firms, caught in out-of-equilibrium conditions engendered by an increase of the demand of their
products, are actually able to implement a creative reaction, as opposed to an adaptive one, depends upon the amount of
pecuniary knowledge externalities that are available at each point in time. The access conditions to external knowledge
are determined by the characteristics of the system, of which technological knowledge and the ensuing technological
change are emergent properties. External knowledge that results from the system’s dynamics is, in turn, an essential
input in the generation of brand new knowledge, in the transformation of the already existing one and in the eventual,
yet crucial, introduction of new technologies. User-producer interactions and transactions play a crucial role as carriers
8
of external knowledge; a role that is stronger, the more advanced are the downstream customers. This context provides
the opportunity to investigate whether knowledge externalities are actually local or global. According to a large
literature, proximity and specifically spatial proximity plays a crucial role in making knowledge spillovers actually
accessible at low absorption costs: hence knowledge externalities would be mainly local. 1 Some evidence about the
effective role of international technology transfer, however, supports the alternative view that knowledge externalities
have a strong global character (Montobbio and Kataischi, 2015).
Along these lines of reasoning it becomes clear that, at the microeconomic level of analysis, what matters in
assuring innovation-based economic progress of the system at large is the derived demand of innovative customers,
rather than the final demand of generic customers. This leads to articulate and revise the demand pull hypothesis
stressing the complementary role of the increase of the derived demand cum the provision of knowledge externalities.
The latter are activated by the market transactions cum knowledge interactions that take place along the vertical filiéres
in the inter-industrial relations.
In conclusion, the competent demand pull hypothesis articulates the view that the increase of demand can
actually pull the eventual introduction of technological innovations, stirring the creative reaction of upstream firms,
only when it is actually combined with the access to external knowledge, available and absorbable at low costs. Firms
within an upstream sector, facing the specific increase of the derived demand for their intermediate products, channeled
by the downstream relations with their customers, can innovate as long as they can actually rely upon pecuniary
knowledge externalities stemming from the technological advance of their competent users. Knowledge transactionscum-interactions with advanced downstream users play a central role in the provision of and access to external
knowledge, which is indispensable in the generation of new technological knowledge and the eventual introduction of
new technologies.
3. Empirical analysis
Keeping in mind the conceptual framework put forward in the previous section, input output tables provide an adequate
framework to test the joint role of user-producer transactions-cum-interactions and the competent demand pull
hypothesis. Thanks to the extended data availability of input output tables for the European economy, we construct the
1
For recent evidence on the local character of knowledge spillovers, see Gehringer (2014).
9
largest panel possible to provide a truly European evidence of the underlying hypothesis. This results in including the
sector-level data on 19 industrial sectors, in 15 EU countries, over the pre-crisis period 1995-2007. 2
The importance to investigate the European economy is straightforward. A lot of effort has been done in the
economic policy of the European Union within the attempts to direct the ongoing process of economic integration to
strengthen the intersectoral linkages and enhance positive spillovers based on innovative activities. 3 In this respect,
input output methodology applied to the European economy seems to be the most appropriate empirical framework of
analysis to grasp the effects of vertical knowledge interactions that take place along with and because of user-producer
transactions (Lundvall, 1985; Von Hippel, 1988; Hauknes and Knell, 2009; Gehringer, 2011a, 2011b, 2012). They enter
into the details of the effects of the demand-side pecuniary knowledge externalities channeled by the downstream
derived demand of innovative users. Moreover, they permit to account for the supply side pecuniary knowledge
externalities channeled from the upstream innovative producers to their downstream users.
The empirical analysis of the demand pull hypothesis has to tackle a central methodological problem that
consists in the identification of the ‘true’ demand effects as distinct from the increase in equilibrium quantity stemming
from the cost reductions. Ex-post, both the upward shift of the demand curve and/or the downward shift of the supply
curve, engendered by increased cost efficiency, can cause an increase of the equilibrium output. The demand pull
hypothesis concerns the first case, yet it seems not trivial to disentangle this effect from the downward shifts of the
supply curve. It is clear that the discovery of an empirical validation of a relationship between the levels of innovative
activity and the increase of the equilibrium demand may be tautological. The levels of innovative activity would be the
cause rather than the consequence of the increase of the equilibrium demand, leading to the obvious endogeneity
concerns. Much empirical investigations aware of the problem elaborated statistical procedures based upon time lags
between the dependent and explanatory variables. The growing evidence about the persistence of innovative activity,
2
The full list of sectors included in the analysis is included in Table A.1 in Appendix A. Countries included in the analysis comprise
the ‘old’ EU member states (Austria, Belgium, Denmark, Finland, France, Germany, Italy, the Netherlands, Portugal, Spain, Sweden,
the UK), in addition to a few Eastern European members from the 2004 wave of the EU enlargement (Czech Republic, Hungary and
Slovak Republic). Due to limited data availability at the sector-level for the remaining Central and Eastern European (CEE)
countries, we had to exclude them from our analysis. Much better data availability for the CEE region is at the macro-level. See, for
instance, Kravtsova and Radosevic (2012) for an investigation regarding the efficiency of national systems of innovation in Eastern
Europe, as well as Kutan and Yigit (2009) for an analysis on the determinants of productivity growth in 8 ‘new’ EU members.
3
Such efforts are formally reflected, for instance and inter alia, in ‘Europe 2020’, an agenda describing the EU’s growth strategy.
Among the crucial policy highlights, industrial competitiveness and industrial innovation are conceived as the driving forces of the
economic performance in Europe.
10
however, undermines the statistical reliance of these procedures (Antonelli et al., 2013). The evidence on the
persistence of innovative activity, in fact, implies that the levels of innovative activity at time (t+n) are strongly
correlated with the levels of innovative activity at time t. To face those problems, we apply two kinds of measures.
First, we concentrate on the contemporaneous demand pulling effects, in the study of which we apply dynamic panel
techniques based on the system GMM methodology. This technique, developed in a microeconomic framework of
analysis, deals with the endogeneity concerns by instrumenting the variables in levels with their first differences and the
variables in first differences with their lagged levels. Second, as our dependent variable, we use not the levels but the
growth rates of TFP, where persistence should be less of an issue. Related to this, because we do not detect
autoregressive process by including the lagged dependent variable, to corroborate our results from the dynamic model,
we apply a static specification in the fixed effects technique.
3.1 Data and methodology
The empirical investigation is based on an unbalanced panel of industry-level yearly observations related to 15 EU
countries. The time span ranges between 1995 and 2007 – the longest period available from the applied data sources.
The data necessary to construct sector-level TFP growth and unit wage come from OECD STAN database.
Annual input output tables come from the World Input-Output Database (WIOD) project. Input output data were used
to calculate the coefficients expressing intersectoral relations – as explained below – in addition to the sector level trade
openness variable and the final demand variable. More precisely, the trade openness variable is obtained for each
downstream sector j as a ratio between the sum of the sector’s imports and exports to this sector’s total output.
TFP methodology
Following Jorgenson and Griliches (1967) and Jorgenson et al. (1987), and assuming the standard growth accounting
framework with a constant returns Cobb-Douglas production function, the logarithmic growth rate of TFP is defined as
𝐾
𝐿
𝑋
𝛥lnTFP𝑖𝑘𝑡 = 𝛥ln𝑌𝑖𝑘𝑡 − 𝛼̅𝑖𝑘𝑡
𝛥ln𝐾𝑖𝑘𝑡 – 𝛼̅𝑖𝑘𝑡
𝛥ln𝐿𝑖𝑘𝑡 – 𝛼̅𝑖𝑘𝑡
𝛥ln𝑋𝑖𝑘𝑡
(1)
where Y is total output of sector i in country k at time t, K is sectoral capital stock, L is labor force measured in terms of
𝑓
total employment and X expresses intermediate inputs. Moreover, 𝛼̅𝑖𝑘𝑡 , where f = (K, L, X), denotes the two-period
average share of factor f over the nominal output, defined as follows:
𝑓
𝑓
𝛼̅𝑖𝑘𝑡
𝑓
(𝛼
+ 𝛼𝑖𝑘𝑡 )⁄
= 𝑖𝑘(𝑡−1)
2
(2)
𝑋𝑖𝑘𝑡
⁄𝑌 and
𝑖𝑘𝑡
(3)
whereas
𝐿
𝛼𝑖𝑘𝑡
=
𝐿𝑖𝑘𝑡
⁄𝑌 ;
𝑖𝑘𝑡
𝑋
𝛼𝑖𝑘𝑡
=
11
𝐾
𝐿
𝑋
𝛼𝑖𝑘𝑡
= 1 − 𝛼𝑖𝑘𝑡
− 𝛼𝑖𝑘𝑡
.
A common practice in the literature using a measure of TFP as a proxy of productivity, is to assume a constant
share of capital over output, α=1/3. We consider such an assumption as a serious drawback, shading the intrinsic
dynamics of industries, countries and over time. For that reason, we depart from the simplifying assumption of a
constant α and adopt a variable measure, reflecting better the changing sectoral, country- and time-specific conditions.4
Input output methodology
In order to exploit market based inter-sectoral linkages and their influence on the sector-level productivity growth, input
output framework has been implemented. We use the annual national input output tables from World Input-Output
Database (WIOD), which are industry per industry tables derived following the Eurostat Manual for Supply, Use, InputOutput Tables methodology (Eurostat, 2008). In the present analysis, given that we want to keep the most disaggregated
structure of sector-level observations, we concentrate exclusively on domestic input output relations. This shouldn’t
undermine the fact that – in some economic and geographic contexts – global knowledge externalities would play a
role, although the recent evidence provided by Gehringer (2014) shows that, in the European context, knowledge
externalities are primarily local. This outcome is also in line with the previous evidence by Branstetter (2001) who
confirms the importance of the so called ‘national component’ in the system of knowledge spillovers. To face the issue,
in our main estimations we control for the possible influence coming from international relations. Our approach here is
twofold. First, in the first set of estimations a variable measuring the intensity of cross-border flows between each
national sector and abroad is included as a separate covariate. This approach permits us to separately assess the
magnitude of purely domestic inter-sectoral relations. In the second step, a new explanatory variable is constructed for
each pair of sectors, such that the influence of trade openness is incorporated directly into the inter-sectoral relation.
Let’s consider a simplified structure of a symmetric input output matrix referred to a country k at time t,
with (𝑠1 , … , 𝑠𝑛 ) sectors, where n is the total number of sectors (Table1).
>> Table 1 here<<
The central part of the matrix describes inter-sectoral linkages based on intermediate inputs’ transactions. The
last row represents the total output produced by each sector - those same values are encountered in the last column
describing total uses of sectoral output. Finally, the column of final demand comprises the sum of the alternative uses
made in final consumption, fixed capital formation and exports.
In the sense of columns, a generic sector j is a producing sector, using x.j inputs from the other upstream sectors
and generating a total output, yj. In the sense of rows, a generic sector i is a supplying sector of the total quantity, yi,
4
There is also a consistent body of literature discussing the use of TFP as an adequate measure of productivity. For an overview, see
Gehringer (2011b).
12
which will be dedicated to the intermediate uses, xi., by each of the column sector, and to the final demand, di. On the
main diagonal, where i=j, intra-sectoral relations can be observed. For instance, x1,1 seen from the input perspective (in
the sense of columns), stays for intermediate inputs that sector 1 receives from itself and uses as intermediate input in
the production process. Instead, seen from the output perspective (in the sense of rows), it represents the quantity of
intermediates offered by sector 1 to that same sector 1.
From the vertical entries, expressing the intermediate needs of each producing column sector j, and considering
additionally the last row of sector-level output, one can obtain a matrix of technical coefficients, A, where the single
component, aij, represents the fraction of inputs received from the supplying sector i over the total output produced by
sector j. Analytically this can be written as
𝑎𝑖𝑗 =
𝑥𝑖𝑗
𝑦𝑗
(4)
Coefficient aij expresses units of direct intermediate requirements delivered by sector i that are necessary to generate 1
unit of total output by sector j.
From matrix A, an inverse Leontief matrix, L, can be obtained through the following transformation:
𝐋 = (I - A)−1
(5)
where I is the identity matrix. A generic element of the inverse Leontief matrix, lij, expresses direct and indirect
requirements demanded by sector j from sector i in terms of intermediate inputs that are necessary to obtain 1 unit of
final demand referred to sector j.
Let’s define matrix B as the matrix of the absolute volumes of direct and indirect intermediate requirements.
This matrix is obtained from the inverse Leontief matrix by multiplying its single row by the corresponding column of
values of output offered by all the sectors to the economy at large. To give an example, considering textile sector, each
element of the column of matrix B corresponding to this sector is a product of the corresponding coefficient of the
Leontief matrix and the value of output produced by the textile sector. This column describes the values of direct and
indirect intermediate requirements necessary in the entire production process of the textile sector. In analytical terms,
matrix B is obtained as:
𝐁 = 𝐋∙𝐘
(6)
where L is the inverse Leontief matrix and Y is the matrix of sectoral outputs.5 A generic component of B, bij, refers to
all inputs coming directly or indirectly from sector i that are necessary to obtain y units of output by sector j. In the
sense of columns, thus, we can read the values of intermediate inputs that are directly and indirectly required by a
column sector j in order to supply the total value of output obtained by this sector. Instead, in the sense of rows, one can
5
The analytical details regarding the calculation of matrix B are illustrated in Appendix B.
13
read the values of intermediate inputs that are directly and/or indirectly produced by a row sector i and that are
demanded by the column sectors in the production of their respective values of output.6
Finally, considering matrix B and defining matrix T as a matrix, with all columns being the same, in the sense
that each collects the TFP growth rate of all the sectors, let’s define two more matrices, R and S.7
𝐑 = [𝐁 ∙ 𝐓]′
(7)
[𝐁 ′
(8)
𝐒=
′
∙ T]
In the sense of columns, the elements r.j of matrix R express direct and indirect intermediate inputs requirements that
this column sector j receives from all the other demanding sectors, but taking into account productivity growth effect of
each demanding sector. The two elements compounding each vertical entry of the matrix fully describe our main
hypothesis: the market transaction element, given by bij, describes the competent demand for intermediate goods
referred to the upstream sector and the technological element, given by the growth rate of TFP of the downstream
sector, through which knowledge externalities should occur. Consequently, matrix R constitutes the crucial element to
measure the competent, demand-driven, influence on the productivity growth of each single upstream sector i. This
demand-driven effect should, thus, come not as a pure intermediate demand effect, yet as an efficiency-enhanced
demand effect, stemming not uniquely from the downstream sector’s demand for intermediate inputs, but also, and
crucially, from an increased total factor productivity of this same demanding downstream sector.
Instead, each column in matrix S, referred to a single downstream user of intermediate inputs, reports direct and
indirect requirements supplied by all its upstream producers, but, again, accounting for the efficiency effect, as
measured by ΔlnTFPi, that a downstream using sector experiences when demanding intermediate inputs from its
upstream producers. This matrix, thus, accounts for efficiency-enhanced supply effect.
6
In the applied work, it is common to post-multiply the inverse Leontief matrix with the vector of final demand or of some
components of it, like exports. Analogously, however, extensions can be made to other concepts as well, like imports, value added,
R&D expenditures (Hauknes and Knell, 2009) or total output, as in our case. The logic driving the construction of matrix B can be
found in the need to match the sectoral productivity growth with the overall production outcomes and not with only final demand (or
the components thereof). It should be, however, noted that with this approach double counting occurs. This is due to the fact that
through the Leontief the entire output of production is disentangled (production inputs and the underlying value added) rather than
the pure effect of final demand, with the value of production is accounted for only once. We are grateful to our anonymous referee
for the very useful comments on these issues.
7
Appendix B contains analytical details concerning the calculation of matrices R and S.
14
3.2 Econometric specification
In line with our theoretical discussion, the focus of the forthcoming empirical investigation is on the quantification of
user-producer market interactions and their influence on sector-level productivity growth. The peculiar feature of our
specification consists in creating a unified framework, where we focus on the demand pulling forces (through the
elements of matrix R), but at the same time we account for the average supply pushing influence (obtained from matrix
S). Consequently, we aim to explain innovativeness of sectors that on their own qualify as producers, when they offer
intermediates to the other demanding sectors, and as users, when they receive inputs from their upstream suppliers.
Accordingly, among explanatory variables, we distinguish between disaggregated, single user-driven (demand) and
averaged producer-driven (supply) factors. The common feature of both types of factors is that they are catalyzed by the
operating of vertical pecuniary knowledge externalities in the sense that both demand-side and supply-side influences
are qualified by innovative capacities of users and producers, respectively.
The econometric model reflecting the key hypothesis that innovation of each agent (sector) is pulled by the
demand of downstream innovative users assumes the following expression:
∆lnTFP𝑗𝑘𝑡 = 𝛽1 + 𝛽2 ∆lnTFP𝑗𝑘𝑡−1 + R𝐣𝐤𝐭 𝛃′3 + 𝛽4 𝑎𝑣𝑠𝑢𝑝𝑗𝑘𝑡 + 𝐙𝐣𝐤𝐭 𝛃′5 + 𝛾𝑘 +𝛿𝑗 + 𝜇𝑡 + 𝜀𝑗𝑘𝑡
(9)
where 𝛽1 is a constant, ∆lnTFP𝑗𝑘𝑡−1 is the lagged dependent variable, 𝛾𝑘 , 𝛿𝑗 and 𝜇𝑡 are the country-, sector- and timespecific effects, respectively, and 𝜀𝑗𝑘𝑡 is the idiosyncratic error term. Vector 𝐑 𝐣𝐤𝐭 corresponds to column j of matrix R at
time t and includes explanatory variables measuring the demand-side influence generated by the downstream sectors. In
the estimation of equation (9), we included 19 sectors, corresponding to 12 manufacturing in addition to 7 market
service sectors. Consequently, agriculture, mining and quarrying, as well as non-market services were not considered.
This is mainly because of their relatively poorer performance in terms of TFP growth, as compared with other sectors. 8
Additionally, in formulating the estimating equation, we were aiming at limiting the risk of over-specification deriving
from the inclusion of too many sectors. In this sense, we concentrated on the impact coming from manufacturing and
market services, based on the presumption that the efficiency-driven impact of demand coming from the excluded
sectors is rather marginal. Consequently, we expect that the cost of such exclusion is less considerable than drawbacks
deriving from over-specification. At the same time, we considered necessary the inclusion of both manufacturing and
service sectors in a unified framework. The motivation of this choice derives mainly from the fact that a selective
8
These sectors were performing at negative TFP growth rates: mining and quarrying (-0.008), public administration (-0.005),
education services (-0.002), health services (-0.004), and other social services (-0.010). An important exception here is agriculture,
for which the growth rate of TFP on average for our sample was of 0.006. Therefore, we perform a robustness check by including
agriculture as well.
15
analysis of only manufactures or only services could lead to an estimation bias. This is because sectoral productivity
growth is supposed to come as a result of interactions with both manufacturing and service sectors. Consequently, the
inclusion of only either of the group could lead to overestimated effects.
Variable avsupjkt is obtained as an average of column j from matrix S and expresses the average supply-side
influence on the productivity growth of sector j deriving from all linkages that this sector maintains with its suppliers by
means of intermediate inputs transactions. 9
Among the control variables in vector Zjkt, we include sector-specific information regarding the rate of change in
sector-level unit wages (wage), final demand (findem) directed towards sector j, and the degree of trade openness (open)
of each single sector j. The inclusion of the wage variable is justified by the presumption that the positive/negative TFP
dynamics could be driven by the wage impulses of the same sign. The variable measuring sector’s j final demand refers
to the hypothesis that the demand-side influence on the sectoral TFP growth could come not only from the intermediate
part, but could be driven by the final consumption as well. Finally, the trade openness variable should account for the
possible influence on sector’s j productivity growth coming from its relations with the sectors located abroad.
It is crucial to observe that the specification of equation (9) and most importantly the use of the information
retrieved from input output tables permits to qualify with clearness the exact role played by the sectors within the
vertical filiére, both on the right- and left-hand side of the equation. More precisely, considering the variables contained
in vector Rjkt, expressing the dynamically efficient demand-side impact, they refer to the transactions based on
competent demand of intermediate inputs coming from downstream sectors and directed towards the supplying
upstream sector. In the same equation, the left-hand side sector might be seen in the quality of a downstream sector
being influenced by an average supply-side effect, as measured by avsupjkt. This means that equation (9) is able to
effectively describe and at the same time distinguish all forward and backward relations between vertically integrated
sectors. This, indeed, is possible thanks to the input output transactions that describe the direction of interaction
between sectors. According to those tables, each sector could and - in most of the cases - actually is user and supplier
with respect to each other sector, in addition to being user and supplier of its own.
The variables contained in vector Rjkt are accounting for the inter-sectoral relationship between two domestic
sectors and the influence of the international relations is captured in a separate covariate expressing the degree of
9
Given the focus of the present paper, we measure demand-side influences coming separately from each downstream sector, whereas
supply-side effects are accounted for by means of an average effect. This is not to say that we consider supply-side effects less
relevant than the demand-side effects. However, given the interest of our analysis in disentangling the manifold impact of demand,
the inclusion of too many explanatory variables expressing sector-specific supply-side effects would impose the risk on the
robustness of our specification.
16
openness to trade of the upstream sector j. Another, more straightforward way to account for the influence of the
‘international component’ is to construct a new vector of bilateral inter-sectoral linkage variables, where the relations
with abroad are permitted to exercise a direct impact on the underlying domestic relation. More precisely, the
components of vector Rjkt become now a product of three factors, the two original ones and the third one measuring the
degree of trade openness of sector j. . The literature documents a lively debate whether knowledge externalities are
intrinsically local because of the crucial role of spatial proximity (Branstetter, 2001) or global because of the
opportunities provided by trade-related user-producer interactions to feed generative interactions at distance (Keller,
1997 and 2009). It is plausible to expect that for some sectors being involved in international relations knowledge
externalities do exert a positive contribution, so that the advantage from the domestic relation would be enhanced. In
this case, the estimated parameter of the variable that includes international openness is expected to be significantly
larger than the parameter of the variable without the international openness. On the other hand, it is also possible that
for some sectors the impact of the involvement in international markets would translate into less intensive knowledge
interactions and thus weaker impact on sectoral productivity growth. In this case, the estimated parameter of the
variable that includes international openness is expected to be lower and less significant than the parameter of the
variable without the international openness.
Table 2 refers to summary statistics of the variables included in the estimation.
>> Table 2 here<<
We first run equation (9) based on the fixed effect model. Nevertheless, this methodology still leaves unresolved
the crucial concern of endogeneity, deriving most importantly from the bi-directional influence between the dependent
and explanatory variables. Indeed, it cannot be excluded that the change in TFP of sector j would influence the
contemporaneous verifications of the variables expressing the demand choice of all the other sectors acquiring
intermediate inputs from j. Moreover, the country and sector specific effects could exercise an influence on the other
explanatory variables as well. For that reason, we follow the approach proposed by Arellano and Bover (1995) and
Blundell and Bond (1998) and we estimate equation (9) with the system GMM method. This method is based on a
system built upon two specifications. The first one is a difference equation, in which variables expressed as firstdifferences are instrumented with theirs lagged levels. The second one is the equation with variables in levels
instrumented with their own lagged first-differences. To be valid, the following assumptions need to hold:
𝑔
𝐸[∆lnTFP𝑗𝑘1 𝜀𝑗𝑘𝑡 ]=𝐸[𝐑 𝐣𝟏 𝜀𝑗𝑘𝑡 ]=𝐸[𝒁ℎ𝟏𝒕 𝜀𝑗𝑘𝑡 ]=0, ∀𝑔, ℎ and t= 2, …, T
𝑔
𝐸[∆∆lnTFP𝑗𝑘2 (𝛾𝑘 + 𝛿𝑗 )]=𝐸[∆𝐃𝐣𝟐 (𝛾𝑘 + 𝛿𝑗 )]=𝐸[∆𝒁ℎ𝟐𝒕 (𝛾𝑘 + 𝛿𝑗 )]=0, ∀𝑔, ℎ and t= 2, …, T
17
(10)
(11)
where 𝑔 ∈ {1 … 𝐺} is the number of variables in vector Rjkt and ℎ ∈ {1 … 𝐻} is the number of control variables in vector
Zjkt.
3.3 Regression results
In Table 3 we report the results of the estimations of equation (9). In column (1), we include the results from the fixed
effect model, whereas in column (2) are the results from the dynamic system GMM estimation procedure. Since,
however, we detect no autoregressive process in the dependent variable, in our discussion, we concentrate on the fixed
effect method.
The results in Table 3 report positive effects on sectoral productivity growth in the case of manufacturing
sectors, in particular, for textiles and textile products, wood and products of wood, paper products, rubber and plastic
products, other non metallic mineral products, machinery and equipment and manufacturing nec. This confirms the
evidence of demand-driven positive productivity effects. It is crucial to underline that the positive impact of
intermediate demand from those sectors on TFP growth of their suppliers is generated along with efficient industry
developments internal to the structure of each user.
Providing a more articulated intuition to the results obtained, the past evidence confirms that textile and clothing
industry in Europe experienced major structural changes in the late 90s, with inefficient plants closing their activities.
Even if the sector is often classified as low-tech considering the intensity of the R&D activities, thanks to the positive
restructuring activities in the past, productivity growth was regarding the entire intra-sectoral chain and some radical
innovations occurred recently. Moreover, it has to be recognized that the activity of textile producers is often placed
within an industrial district where inter-sectoral linkages with both upstream and downstream producers acquire a
peculiar meaning. In such an environment, pecuniary knowledge externalities at play in the market transactions between
users and producers constitute an integral part of the landscape and tend to be more incident than in a standard nonclustered industrial structure. The recalled evidence confirms the role of the sector as an innovative user in contributing
to upstream productivity growth.
At the same time, the downstream process of development of innovative machinery and equipment is reflected in
the creative adaptation of their upstream suppliers, committed to maintain their market position and thus willing to
remain up to date with the dynamic performance of machinery and equipment producers. The loop of circular feedback
is especially clear here with downstream producers introducing product innovations and pulling the supply of new
products from their upstream producers so as to introduce subsequent process innovations downstream. In turn,
upstream producers of the machinery innovate because they need to face positively the pulling challenges and
opportunities provided by the growth of their downstream customers.
18
>> Table 3 here <<
The case of machinery and equipment is especially relevant. The sector supplies capital goods that embody
product innovations, whose adoption feeds process innovations in downstream users. This explains its role as a strategic
supplier for many other sectors. Moreover, firms producing machinery and equipment can develop product innovations
in response to the specific technical suggestions of their customers as channeled by fertile user-producer interactions
(Von Hippel, 1976).
The strong evidence regarding rubber and plastic products reflects the focus of European industrial policy on the
environmental issues. The intensive stimulus provided for the development of innovative products and recycling
processes, complying with the environmental protection, led rubber and plastic producers to successful waves of ecoinnovations.10 Moreover, being downstream producers of rubber and plastic in continuous interaction with their
suppliers - among which doubtlessly also the producers of rubber and plastic machinery - they managed to activate a
dynamic process of transactions-cum-interactions with an evident positive impact on sector-level total factor
productivity growth upstream.
Already from the previous discussion it appears with clearness that the innovative process is nested in bidirectional user-producer relationships. The importance of this phenomenon is confirmed also in our results, as the
average supply-push effect (avsup) appeared to be significantly positive as well. 11 This evidence of a significant average
effect confirms the findings by Gehringer (2011a, 2011b and 2012) who offers a detailed investigation aiming to reveal
empirically the importance of supply-side pecuniary knowledge externalities in the European economies. Finally, the
findings confirm the simultaneous role of demand and supply side factors, in line with the arguments of Mowery and
Rosenberg (1979) and Arthur (2007).
Instead, negative evidence has been reported for two service sectors, construction and real estate. Regarding
construction sector, a number of past investigations suggested that (labour) productivity in construction declined,
10
New technological process for rubber and plastic recycling enables the complete decomposition of rubber and plastic waste to the
commercial components. Similarly, the environmental concerns brought dynamic developments in the automotive sector, with the
consequent need of high-performance plastic materials, like, for instance, high temperature-resistance plastics.
11
Due to a strongly significant result regarding the supply-side influence, in a separate estimation procedure, not reported here, we
excluded avsup from the estimation. The results relative to the demand-side influence appeared to be stronger than before, justifying
the need to consider a joint influence of the supply- and demand-side in a unique specification.
19
whereas an opposite tendency could be contemporaneously observed for the majority of manufacturing sectors. 12 This
finding, put together with the evidence of the recent overheating of the activity in the European construction sector in
European economies, with the most problematic cases of Spain and Ireland, partly seems to support our results.
Moreover, rising employment in construction sector was probably linked with unsustainable real estate booms.
Nevertheless, given peculiarity of the effects at play in our research framework, further investigation of economic
forces driving such effect would be needed to disentangle its more precise nature.
Finally, controlling for the incidence of the final demand (findem) and of the degree of sector-level trade
openness (open) produced insignificant results in the case of both methods. The impact of unit wages (wage) was
positive, though only in the case of the fixed effect model.
The fact that the trade openness variable produced insignificant result shouldn’t immediately exclude any role of
the international context. Indeed, in estimations reported in Table 4, we include in matrix R variables directly
accounting for the impact of the international trade.
>> Table 4 here <<
To better interpret and compare the results of these estimations with the regressions reported in Table 3,
standardized coefficients relative to the FE estimations have been calculated and included in Table 5.
>> Table 5 here <<
The evidence provided by table 5 is quite important. The comparison between the results of Table 3 where
equation 9 includes the independent variable without the degree of international trade openness and Table 4 where the
independent variable includes the degree of international trade openness show that knowledge externalities have a
strong and significant global effect. The parameter of the variable that includes the degree of international openness is
significantly stronger in most industries: food, textile, paper and pulp, chemistry, rubber, other non-metallic minerals,
machinery, electrical machinery, transportation equipment and services. The variable that does not include the degree of
international openness is never higher and/or more significant than the variable constructed with the degree of
international trade openness. These results seem to provide strong support to the hypothesis that knowledge externalities
have a strong global character: countries do benefit from the trade-related user-producer knowledge interactions that
take place across borders. Institutional proximity matters as well as spatial proximity.
12
Those studies refer basically to the US data before 2000, as for instance the analysis by Teicholz (2001). The evidence regarding
the European economies is missing. Nevertheless, a consensus among authors studying construction sector is that the complexity of
the operating in construction sector makes it all the more difficult to perform at a level-playing field. For a survey of studies
analyzing structural issues in construction sector, see Dubois and Gadde (2001).
20
Although our results are useful to formulate precise policy options at the common, European level, we cannot
aim at delivering policy implications, focusing on one type of sectoral activity or another in a specific country or region.
This is most importantly due to the fact that we obtained general European evidence on the demand pulling the
innovativeness. Indeed, the geographic composition of economic activity matters in influencing innovation (Bottazzi
and Peri, 2003). Further investigation in a single country perspective is needed to deliver more concrete, country-level,
innovation policy implications.
3.4. Robustness checks
In considering the efficiency-driven demand effects on productivity growth in the previous section it was essential to
measure the impact of derived demand conditioned on improved innovativeness of downstream users. Consequently, in
the previous investigation, the explanatory variables in matrix R have been constructed as compound effects of two
competing forces, the market transaction component (represented by the coefficient of the inverse Leontief matrix) and
the productivity component (in terms of TFP growth). However, a justified expectation could lead to argue that the
evidence of any significant effect would be driven by either of those forces. It is, thus, reasonable to empirically check
for this possibility. To this end, we run further two specifications, one measuring the possible pure demand effect and
another one trying to detect the pure technological influence.
>> Table 6 here <<
More precisely, the first estimation consists in replacing vector Rjkt of variables expressing the efficiencyenhanced demand effect with vector Bjkt corresponding to column j of the transpose of matrix B. Each element of this
column refers to direct and indirect intermediate inputs that the upstream sector delivers to each of the downstream
sectors. In such a way, we want to take into account the pure effect on upstream productivity growth coming from
sectoral demand for intermediate goods, separated from the productivity effect of the demanding sector. Instead, the
second estimation, in analogy to the previous one, consists in replacing vector Rjkt with the vector of sectoral TFP
growth of the 19 sectors taken under investigation. Recalling the discussion offered in Section 2 regarding the
distinction between pecuniary knowledge externalities and pure technological externalities, this second specification is
of a great importance, as it verifies the existence (or the absence) of pure technological externalities, to which a great
role has been assigned in the past theoretical and empirical literature (Griliches, 1979).
The results in Table 6 report almost no evidence from operating of pure effects both in terms of intermediate
demand (column (1)) and in terms of inter-sectoral productivity growth spillovers (column (2)). This lack of evidence is
crucial for the findings of the previous section, as it strengthens the hypothesis sustaining that an indispensable element
for the demand-pull influence on productivity of producers to occur is given by efficiency of downstream users. This
21
also supports the hypothesis that the assumption of knowledge interactions-sine-transactions, typical for pure
technological externalities, is not supported with empirical evidence (Gehringer, 2011a).
A second set of robustness checks refers to the sectoral composition chosen for the empirical analysis. In the
previous section, we have excluded agriculture, mining and some of the service sectors, based on the expectation that
the evidence of efficiency-driven demand pulling originated from these sectors wouldn’t be strong. This was motivated
by the relatively weak performance of these sectors in terms of productivity growth. Nevertheless, agriculture presents
an important exception in this respect, as it experienced - on average for the observed sample - positive growth rate of
productivity. Consequently, we rerun our regressions by including agricultural sector. The results presented in column
(1) of Table 7 are in line with these presented in Table 3. Additionally, the coefficient on agriculture turns to be positive
and significant, confirming that the specialized demand coming from this sector contributed to positive productivity
dynamics of the upstream sectors.
Finally, given that from our baseline results we could observe a relative weaker role played by service sectors in
sustaining the demand-driven technological influences, we performed additional estimations by concentrating
exclusively on manufacturing sectors. In this way, we want to check whether the inclusion of too many covariates
wouldn’t undermine the estimation results referring to sectors playing a positive and significant role in the system of
intersectoral relations. The results presented in column (2) of Table 7 confirm the strong impact exercised by the same
manufacturing sectors already identified in the previous regressions. Moreover, the magnitude of the sectoral influences
remains robust to the exclusion of service sectors.
>> Table 7 here <<
4. Conclusions
The identification of vertical knowledge externalities, based upon the knowledge interactions that take place within the
context of user-producer transactions and the appreciation of their central role in the generation of new technological
knowledge, provide new foundations to implement and articulate the intuition of Nicholas Kaldor that public
intervention can become a structural component of economic policy when and if it consists of a support of aggregate
demand, able to pull technological change. The contribution of Jacob Schmookler helped refining the Kaldorian
hypothesis, focusing attention on the role of private derived demand for capital and intermediary inputs and selected
investments. Nevertheless, further implementation of the demand pull hypothesis, building upon the KaldorSchmookler line of analysis, requires better understanding of its microeconomic foundations. The new literature on the
economics of knowledge provides useful guidance in these attempts and permits to better grasp the conditions that
make possible the very generation of technological knowledge and the eventual introduction of technological
22
innovations. This implies the scrutiny of the introduction of technological and organizational innovations as systemic
processes, based on the active participation of individual agents to the knowledge commons embedded in the structure
of economic systems and implemented by vertical knowledge interactions. A crucial feature of those interactions is
their being channeled by user-producer transactions, pushing further the appreciation of the endogenous character of
technological change.
In our approach, we elaborated the microeconomic foundations of the competent demand pull hypothesis,
building upon the Schumpeterian approach to innovation as a form of reaction to unexpected changes in the conditions
of factor and product markets, including the levels of the demand, which becomes actually creative when and where
firms can rely upon the access to external sources of technological knowledge. The competent demand pull hypothesis
stresses the complementarity between demand pull and knowledge externalities. More specifically, we identified the
derived demand of innovative customers as the key mechanism of the competent demand pull hypothesis. From an
empirical viewpoint we contributed to the literature on the demand pull hypothesis in two crucial points. First, we filled
a major conceptual gap deriving from the lack of a microeconomic understanding of the pulling mechanism of derived
demand. Second, by taking advantage of the opportunity provided by the implementation of input output tables, we
tested the role of derived demand pull cum vertical knowledge externalities on the upstream rates of growth of total
factor productivity, channeled by user-producer interactions.
The results of our investigation, performed on a panel of 15 European countries in the years 1995-2007, confirm
that the rates of introduction of innovations, as measured by the rates of increase of total factor productivity, are
significantly associated with the increase of the derived demand of downstream sectors associated with their own rates
of increase of total factor productivity. The evidence supports the argument that generic increases of the aggregate
demand are not sufficient to pull the rate of introduction of innovations. Only when the increase of demand is qualified
as it is associated to the actual increase of the rates of innovation introduction of the user sectors, the demand pull
mechanism becomes effective (Peters et al., 2012). Consequently, such evidence of an important role played by
innovative user in their interactions with innovative producers, combined with the quasi-public good character of
knowledge, implies that externalities may be crucial in shaping industrial development in Europe. Additionally,
moreover, we demonstrate that those externalities require a market context and its effective transactions to operate. A
crucial result in this sense is, thus, that technological externalities are not pure yet pecuniary. Finally, our results suggest
that knowledge externalities are both local and global. Agents benefit from the trade-related user-producer knowledge
interactions that take place both within and across borders. Both institutional and spatial proximity matter in
implementing the knowledge interactions that make external knowledge available.
23
The implications of competent demand pull hypothesis are important both from the viewpoint of economics and
for economic policy. From the viewpoint of economic analysis the qualification that demand matters in pulling
innovation when it is associated with stronger user-producer transactions, channeling knowledge interactions, amounts
to suggesting a way to reconcile the Keynesian and the Schumpeterian traditions. Demand matters as in the Keynesian
tradition, articulated by Nicholas Kaldor, only when it is derived demand of competent users that have innovated and
are able to support the innovative efforts of their suppliers. At the same time, the new emphasis of knowledge
externalities that support the creative reaction of firms clearly impinges upon the Schumpeterian tradition. From the
economic policy viewpoint, the results of our analysis have important implications for public support of innovative
initiatives, as they credit the need to identify sophisticated public interventions able to combine the support to the
demand with the exploitation of technological opportunities. In this sense, public support to sustain demand should be
combined with complementary support to innovation policies not only in upstream, but also in downstream users.
The generic support of aggregate demand may speed the diffusion of innovations, as suggested by Kaldor, but is
not likely to pull effectively the generation of new technological knowledge and the introduction of innovations. The
identification of user-producer transactions-cum-knowledge interactions as the systemic mechanisms, by means of
which the participation of each agent to knowledge commons is effectively enhanced, stresses the need to proceed to
the selection of the filiéres along which competent and innovative users can support the increase of the derived demand,
including both the demand for capital goods and for intermediary inputs. This would have the chance to pull the overall
increase in the rate of generation of technological knowledge and the introduction of technological and organizational
innovations.
A competent and specific demand, as opposed to a generic one, based upon both public procurement and private
demand stirred by dedicated regulations, can pull effectively the generation of new technological knowledge and the
introduction of new technologies. This will become effective provided that the competent users are able to identify the
relevant technological opportunities available at each point in time and the structures of vertical transactions that are
actually associated to active knowledge interactions.
Acknowledgments
We are grateful to the Editor, Richard Frensch, and the two anonymous referees for their excellent support in reviewing
the paper. We also thank Giuseppe Scellato and other participants of the Conference “Re-Imagining Europe:
Demand-Driven Innovation and Economic Policy” for very helpful comments to the previous draft of the paper. The
authors acknowledge the financial support of the European Union D.G. Research with the Grant number 266959 to the
research project ‘Policy Incentives for the Creation of Knowledge: Methods and Evidence’ (PICK-ME), within the
24
context of the Cooperation Program / Theme 8 / Socio-economic Sciences and Humanities (SSH), in progress at the
Collegio Carlo Alberto and the University of Torino.
Bibliography
Andersen, E.S., 2007. Innovation and demand, in: Hanusch, H., Pyka, A. (Eds), The Elgar Companion to NeoSchumpeterian Economics, Edward Elgar, Cheltenham, pp. 754-765.
Antonelli, C., 2008. Pecuniary knowledge externalities: The convergence of directed technological change and the
emergence of innovation systems, Ind. Corp. Chang., 17, 1049-1070.
Antonelli, C. (Ed), 2011. Handbook on the Economic Complexity of Technological Change, Edward Elgar,
Cheltenham.
Antonelli, C., 2013. Knowledge governance, pecuniary knowledge externalities and total factor productivity growth,
Econ. Dev. Q., 27, 62-70.
Antonelli, C., Crespi, F., Scellato, G., 2013. Internal and external factors in innovation persistence, Econ. Innov. New
Technol., 22, 256-280.
Antonelli, C., Link, A., (Eds) 2015. Handbook on the Economics of Knowledge, Routledge, London.
Arellano, M., Bover, O., 1995. Another look at the instrumental variables estimation of error-components model, J.
Econom., 68, 29-51.
Arthur, W.B., 2007. The structure of invention, Res. Policy, 36, 274-287.
Arthur W. B., 2009. The Nature of Technology. What It Is and How It Evolves, Free Press, New York.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models, J. Econom., 87,
11-143.
Bottazzi, L., Peri, G., 2003. Innovation and spillovers in regions: Evidence from European patent data, Eur. Econ. Rev.,
47, 687-710.
Branstetter, L.G., 2001. Are knowledge spillovers international or intranational in scope? Microeconomic evidence
from the U.S. and Japan. J. Int. Econ., 53, 53-79.
Brouwer, E., Kleinknecht, A., 1999. Keynes-plus? Effective demand and changes in firm-level R&D: an empirical note,
Camb. J. Econ., 23, 385–91.
Di Stefano, G., Gambardella, A., Verona, G., 2012. Technology push and demand pull perspectives in innovation
studies: Current findings and future research directions, Res. Policy, 41, 1283-1295.
Dosi, G., 1982. Technological paradigms and technological trajectories. A suggested interpretation of the determinants
and directions of technical change, Res. Policy, 11, 147–162.
25
Dubois, A., Gadde, L-E., 2002. The construction industry as a loosely coupled system: Implications for productivity
and innovation, Constr. Manag. Econ., 20, 621-631.
Eurostat, 2008. Eurostat manual of supply, use and input-output tables. Eurostat Methodologies and Working Papers.
Available at: http://epp.eurostat.ec.europa.eu/cache/ITY_OFFPUB/KS-RA-07-013/EN/KS-RA-07-013-EN.PDF.
Fontana, R., Guerzoni, M., 2008. Incentives and uncertainty: An empirical analysis of the impact of demand on
innovation. Camb. J. Econ., 32, 927–946.
Gehringer, A., 2011a. Pecuniary knowledge externalities and innovation: Intersectoral linkages and their effects beyond
technological spillovers, Econ. Innov. New Technol., 20, 495-515.
Gehringer, A., 2011b. Pecuniary knowledge externalities across European countries—Are there leading sectors? Ind.
Innov., 18, 415-436.
Gehringer, A., 2012. A new sectoral taxonomy based on pecuniary knowledge externalities: Knowledge interactions in
a vertically integrated system, Econ. Sys. Res., 24, 35-55.
Gehringer, A., 2014. Knowledge externalities and sectoral interdependences: Evidence from an open economy
perspective. Paper presented at the EARIE Annual Conference 2014, Milan, 29-31 May.
Geroski, P.A., Walters, C.F., 1995. Innovative activity over the business cycle, Econ. J., 105, 916-28.
Griliches, Z., 1979. Issues in assessing the contribution of research and development to productivity growth, Bell J.
Econ., 10, 92–116.
Guerzoni, M., 2010. The impact of market size and users’ sophistication on innovation: The patterns of demand, Econ.
Innov. New Technol., 19, 113-126.
Hauknes, J., Knell, M., 2009. Embodied knowledge and sectoral linkages: An input-output approach to the interaction
of high- and low-tech industries, Res. Policy, 38, 459-469.
Jacobs, J., 1969. The Economy of Cities, Random House, New York.
Jaffe, A.B., 1988. Demand and supply influences in R&D intensity and productivity growth, Rev. Econ. Stat., 70, 431437.
Jorgenson, D.W., Griliches, Z., 1967. The explanation of productivity change, Rev. Econ. Stud., 34, 249-283.
Jorgenson, D.W., Gollop, F.M., Fraumeni, B.M., 1987. Productivity and U.S. Economic Growth, Harvard University
Press, Cambridge, MA.
Kaldor, N., 1966. Causes of the Slow Rate of Growth in the United Kingdom, Cambridge University Press, Cambridge.
Kaldor, N. 1972. The Irrelevance of Equilibrium Economics, Cambridge University Press, Cambridge.
Keller, W. 1997. Are international R&D spillovers trade-related? Analyzing spillovers among randomly matched trade
partners. European Econ. Rev. 42(8), 1469-1481.
26
Keller, W. 2010. International trade, foreign direct investment, and technology spillovers, in: Hall, B.H., Rosenberg, N.
(eds.), The Handbook of the Economics of Innovation, Volume 2. Elsevier, Amsterdam.
Kleinknecht, A., Verspagen, B., 1990. Demand and innovation: Schmookler re-examined, Res. Policy, 19, 387-394.
Kravtsova, V., Radosevic, S., 2012. Are systems of innovation in Eastern Europe efficient? Econ. Sys., 36, 109-126.
Kutan, A.M, Yigit, T.M., 2009. European integration, productivity growth and real convergence: Evidence from the
new member states, Econ. Sys., 33, 127-137.
Lundvall, B.A., 1985. Product Innovation and User-Producer Interaction, Aalborg University Press, Aalborg.
Montobbio, F., Kataishi, R., 2015. The international dissemination of technological knowledge, in: Antonelli, C., Link,
A., (eds), Handbook of the Economics of Knowledge, Routledge, London.
Mowery, D., Rosenberg, N., 1979. The influence of market demand upon innovation: A critical review of some recent
empirical studies, Res. Policy, 8, 102–153.
Nemet, G.F., 2009. Demand-pull, technology-push, and government-led incentives for non-incremental technical
change, Res. Policy, 38, 700-709.
Peters, M., Schneider, M., Griesshaber, T., Hoffmann, V.H., 2012. The impact of technology-push and demand-pull
policies on technical change – Does the locus of policies matter? Res. Policy, 41, 1296-1308.
Piva, M., Vivarelli, M., 2007. Is demand-pulled innovation equally important in different groups of firms? Camb. J
Econ., 31, 691-710.
Scherer, F.M., 1964. The Weapons Acquisition Process: Economic Incentives, Harvard University Graduate School of
Business Administration.
Scherer, F.M., 1982. Demand-pull and technological invention: Schmookler revisited, J Ind. Econ., 30, 225-237.
Schmookler, J., 1966. Invention and Economic Growth, Harvard University Press, Cambridge, MA.
Teicholz, P., 2001. Discussion: U.S. construction labor productivity trends, 1970-1998, J. Constr. Eng. Manag., 126,
427-429.
Von Hippel, E., 1976. The dominant role of users in the scientific instrument innovation process, Res. Policy, 5, 212–
239.
Von Hippel, E., 1988. The Sources of Innovation, Oxford University Press, Oxford.
Von Hippel, E., 1994. “Sticky information” and the locus of problem-solving: implications for innovation, Manag. Sci.,
40, 429–439.
Von Hippel, E., 1998. Economics of product development by users: The impact of “sticky” local information, Manag.
Sci., 44, 629–644.
Von Hippel, E., 2005. Democratizing Innovation, The MIT Press, Cambridge, MA.
27
Walsh, V., 1984. Invention and innovation in the chemical industry: Demand-pull or discovery-push? Res. Policy, 13,
211-34.
Weitzman, M.L., 1996. Hybridizing growth theory, Am. Econ. Rev., 86, 207-212.
Weitzman,
M.L.,
1998.
Recombinant
growth,
28
Q.
J.
Econ.,
113,
331-360.
Table 1
A simplified representation of the input output framework
sector/sector
s1
sj
sn
final demand
output
s1
x1,1
x1,j
x1,n
d1
y1
si
xi,1
xi,j
xi,n
di
yi
sn
xn,1
xn,j
xn,n
dn
yn
output
y1
yj
yn
Table 2
Summary statistics
Variable
Observations
Mean
Standard Dev.
Minimum
Maximum
ΔlnTFP
4520
0.002
0.040
-0.405
0.369
food
4875
-0.001
0.223
-8.414
4.583
text
4875
-0.002
0.095
-1.886
2.389
wood
4875
0.001
0.021
-0.465
0.471
pap
4875
-0.001
0.060
-1.318
1.051
chem
4875
0.009
0.630
-11.538
26.920
rub
4875
0.003
0.052
-0.540
1.106
met
4875
0.007
0.234
-4.158
8.481
mach
4875
0.005
0.109
-2.090
2.123
elec
4875
0.017
0.302
-4.075
8.097
treq
4875
0.022
0.531
-11.179
12.395
manu
4875
-0.002
0.058
-1.373
0.931
util
4875
0.003
0.142
-2.640
4.471
constr
4875
-0.040
0.626
-18.483
10.034
whole
4875
-0.000
0.345
-6.691
7.345
hot
4875
-0.019
0.197
-5.128
3.117
trans
4875
0.016
0.279
-3.625
5.279
fin
4875
0.007
0.279
-5.918
5.491
real
4875
0.066
0.708
-7.657
15.322
avsup
4550
0.000
0.186
-0.890
1.073
wage
4849
32404
41518
858
425114
findem
4875
30607
49825
9
461683
open
4875
0.419
0.358
-0.039
1.559
Table 3 Results from the estimation of equation (9)
FE
l.Δ(TFP)
food
text
wood
0.003
(0.003)
0.020**
(0.007)
0.205***
29
sys GMM
-0.029
(0.047)
0.013***
(0.004)
0.030**
(0.013)
0.214***
pap
chem
rub
onm
met
mach
elec
treq
manu
util
constr
whole
hot
trans
fin
real
avsup
wage
findem
open
observations
R-squared
(0.062)
0.045**
(0.020)
0.002
(0.002)
0.088***
(0.023)
0.115***
(0.025)
0.004
(0.010)
0.026***
(0.008)
0.008
(0.010)
0.002
(0.004)
0.051***
(0.014)
0.018*
(0.010)
-0.008***
(0.002)
-0.008**
(0.003)
0.003
(0.007)
-0.001
(0.003)
0.001
(0.004)
-0.009***
(0.002)
0.034***
(0.005)
0.001**
(0.003)
-0.000
(0.000)
-0.016
(0.011)
4519
0.328
(0.066)
0.048***
(0.014)
0.013***
(0.003)
0.096**
(0.045)
0.117***
(0.024)
0.010
(0.008)
0.036***
(0.010)
0.018*
(0.010)
0.012**
(0.006)
0.062***
(0.014)
0.026**
(0.012)
0.001
(0.004)
0.004
(0.004)
0.016**
(0.006)
0.011
(0.004)
0.010
(0.006)
0.001
(0.002)
0.001
(0.001)
-0.000
(0.000)
-0.000
(0.000)
-0.016
(0.025)
4146
Hansen (p-value)
0.102
AB m-2 (p-value)
0.446
Diff-in-Hansen (p-values)
excluding gmm instr.
0.127
exogeneity of gmm instr.
0.245
Note: ***, ** and * imply significance levels at 1%, 5% and 10%, respectively. Robust standards errors are
included in parentheses. Country, sector and time dummies are included. The models are validated with
appropriate tests: overall R-squared for the fixed effects method and Hansen test of joint validity of instruments
(p-value). Moreover, Arellano-Bond test for no autocorrelation of order 2 and the Difference-in-Hansen test of
exogeneity of instrument subsets have been considered for system GMM. We allow for two-period instruments
of the dependent variable.
Table 4 Results from the estimation of equation (9) with sectoral covariates directly accounting for
the influence of international relations
FE
l.Δ(TFP)
food
text
0.064***
(0.019)
1.468***
30
sys GMM
-0.037
(0.046)
0.064**
(0.035)
1.856***
(0.247)
0.222***
(0.026)
0.347***
(0.051)
0.007***
(0.001)
1.288***
(0.157)
1.068***
(0.154)
0.003
(0.003)
0.054***
(0.009)
0.016***
(0.003)
0.025***
(0.004)
0.153***
(0.022)
0.226***
(0.028)
-0.015***
(0.003)
-0.051**
(0.020)
0.072***
(0.027)
0.005
(0.003)
0.006***
(0.002)
-0.020***
(0.005)
0.026***
(0.001)
0.000*
(0.000)
-0.000
(0.000)
4519
0.353
wood
pap
chem
rub
onm
met
mach
elec
treq
manu
util
constr
whole
hot
trans
fin
real
avsup
wage
findem
observations
R-squared
(0.516)
0.217***
(0.066)
0.348***
(0.088)
0.006
(0.004)
1.178***
(0.359)
1.094***
(0.404)
0.002
(0.008)
0.056***
(0.021)
0.016
(0.012)
0.022
(0.029)
0.148***
(0.048)
0.181**
(0.078)
-0.028**
(0.012)
-0.057
(0.048)
0.110
(0.095)
0.003
(0.005)
0.001
(0.006)
-0.017
(0.014)
0.028***
(0.001)
-0.000
(0.000)
-0.000**
(0.000)
4146
Hansen (p-value)
0.085
AB m-2 (p-value)
0.478
Diff-in-Hansen (p-values)
excluding gmm instr.
0.084
exogeneity of gmm instr.
0.379
Note: ***, ** and * imply significance levels at 1%, 5% and 10%, respectively. Robust standards errors are
included in parentheses. Country, sector and time dummies are included. The models are validated with
appropriate tests: overall R-squared for the fixed effects method and Hansen test of joint validity of instruments
(p-value). Moreover, Arellano-Bond test for no autocorrelation of order 2 and the Difference-in-Hansen test of
exogeneity of instrument subsets have been considered for system GMM. We allow for two-period instruments
of the dependent variable.
Table 5 Standardized coefficients relative to estimations in Table 3 and 4
FE results
Results Table 3
Stand. Coeff.
FE results
Results Table 4
Stand. Coeff.
food
0.003
0.017
0.064***
0.038
text
0.020**
0.048
1.468***
0.071
wood
0.205***
0.108
0.222***
0.106
pap
0.045**
0.068
0.347***
0.076
31
chem
0.002
0.000
0.007***
0.000
rub
0.088***
1.387
1.288***
15.106
onm
0.115***
0.148
1.068***
0.089
met
0.004
0.003
0.003
0.000
mach
0.026***
0.152
0.054***
0.236
elec
0.008
0.022
0.016***
0.023
treq
0.002
0.000
0.025***
0.000
manu
0.051***
0.385
0.153***
0.712
util
0.018*
0.239
0.226***
0.780
constr
-0.008***
-0.012
-0.015***
-0.008
whole
-0.008**
-0.028
-0.051**
-0.020
hot
0.003
0.047
0.072***
0.342
trans
-0.001
0.000
0.005
0.000
fin
0.001
0.009
0.006***
0.005
real
-0.009***
-0.044
-0.020***
-0.010
Note ***, ** and * report the significance level of 1%, 5% and 10%, respectively.
Table 6 Alternative estimations of equation (9) considering the effects of intermediate demand
and the spillover effects of sectoral productivity growth
food
text
wood
pap
chem
rub
onm
met
mach
elec
treq
manu
util
constr
whole
hot
trans
fin
real
avsup
wage
(1)
-0.003
(0.003)
0.0001
(0.001)
-0.001
(0.002)
-0.012*
(0.005)
-0.000
(0.003)
0.001
(0.002)
-0.004
(0.002)
-0.002
(0.002)
0.001
(0.003)
0.000
(0.002)
0.002
(0.002)
-0.003
(0.002)
-0.002
(0.002)
-0.003
(0.004)
-0.000
(0.003)
0.004
(0.004)
-0.008
(0.004)
0.001
(0.003)
0.009
(0.004)
0.003***
(0.004)
0.000
(0.000)
32
(2)
0.029
(0.034)
0.047
(0.029)
0.041*
(0.023)
0.036
(0.028)
0.018
(0.012)
0.041
(0.025)
0.011
(0.031)
-0.012
(0.035)
0.054*
(0.030)
0.029
(0.022)
0.028
(0.027)
0.019
(0.024)
0.004
(0.026)
0.010
(0.026)
-0.027
(0.033)
0.023
(0.026)
0.093**
(0.043)
0.023
(0.019)
0.034
(0.030)
0.030***
(0.004)
0.000
(0.000)
findem
open
observations
R-squared
-0.000
(0.000)
-0.014
(0.011)
4519
0.140
-0.000
(0.000)
-0.013
(0.011)
4519
0.313
Note: ***, ** and * imply significance levels at 1%, 5% and 10%, respectively. Estimations are run
according to the fixed effects method. Robust standards errors are included in parentheses. Country, sector
and time dummies are included.
Table 7 Robustness check on the composition of sectors.
agricult
food
text
wood
pap
chem
rub
onm
met
mach
elec
treq
manu
util
constr
whole
hot
trans
fin
real
avsup
wage
findem
open
observations
R-squared
(1)
0.019***
(0.005)
0.005*
(0.003)
0.022***
(0.008)
0.209***
(0.063)
0.048**
(0.020)
0.005**
(0.002)
0.091***
(0.023)
0.116***
(0.024)
0.006
(0.010)
0.028***
(0.008)
0.010
(0.010)
0.004
(0.004)
0.054***
(0.014)
0.021**
(0.010)
-0.005**
(0.002)
-0.005*
(0.003)
0.006
(0.007)
0.002
(0.003)
0.005
(0.006)
-0.006***
(0.002)
0.027***
(0.005)
0.000***
(0.000)
-0.000
(0.000)
-0.016
(0.011)
4519
0.337
33
(2)
0.009***
(0.002)
0.025***
(0.007)
0.206***
(0.061)
0.049**
(0.020)
0.007***
(0.002)
0.092***
(0.023)
0.111***
(0.024)
0.007
(0.010)
0.030***
(0.008)
0.012
(0.010)
0.007*
(0.004)
0.055***
(0.014)
0.023***
(0.003)
0.000***
(0.000)
-0.000**
(0.000)
-0.014
(0.011)
4519
0.306
Note: ***, ** and * imply significance levels at 1%, 5% and 10%, respectively. Estimations are run
according to the fixed effects method. Robust standards errors are included in parentheses. Country, sector
and time dummies are included.
34